CockroachDB uses a kv store to store rows. For example, a segment with stream_id and position ( s, p) is stored as something like
(metainfo_segments_s_p, root_piece_id_encrypted_key_nonce......inline_data_remote_pieces)
This makes us download all data each time we want an information about a segment, when in many cases we only need stream_id and position.
With column families, we would have as many kv store values per row as column families. For example, if we created a family for inline_data, we would get an additional kv entry which would look like:
(metainfo_segments_s_p_inline_data, inline_data)
This would not be retrieved when we only want stream_id and position.
What we could get from that:
bytes exchanged in the cluster decrease
operation like listing objects could be a bit faster
load better spread across the crdb cluster
decrease crdb cost (price depends mostly CPU/RAM, then data).
CockroachDB uses a kv store to store rows. For example, a segment with stream_id and position
( s, p)is stored as something like(metainfo_segments_s_p, root_piece_id_encrypted_key_nonce......inline_data_remote_pieces)This makes us download all data each time we want an information about a segment, when in many cases we only need
stream_idandposition.With column families, we would have as many kv store values per row as column families. For example, if we created a family for inline_data, we would get an additional kv entry which would look like:
(metainfo_segments_s_p_inline_data, inline_data)This would not be retrieved when we only want stream_id and position.
What we could get from that: